---
title: Parakeet TDT 0.6B v3 on SIE | Superlinked
description: SIE, an open-source inference engine, serves Parakeet TDT 0.6B v3, an open speech model from NVIDIA that turns recorded calls and meetings into text.
canonical_url: https://superlinked.com/models/nvidia-parakeet-tdt-0-6b-v3
last_updated: 2026-11-07
---

# Parakeet TDT 0.6B v3

An open speech model from NVIDIA that turns recorded calls and meetings into text.

Model id: `nvidia/parakeet-tdt-0.6b-v3` · Endpoint: `extract` · [Hugging Face](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3)

Listed under [Speech to text](/models#speech) on [the model index](/models.md).

| Accuracy | Time per request | Serverless price |
|---|---|---|
| – (LibriSpeech, 256 clean clips, word error rate · not measured) | – (Not timed one request at a time) | $0.00075 per audio minute |

## On the task pages

- [Transcribe audio 83% cheaper than GPT-Transcribe](/speech-to-text): $45 Cost per 1,000 audio hours

## Call it

On SIE Serverless:

```python
from pathlib import Path
from sie_sdk import SIEClient

client = SIEClient(api_key="sk-sie-…")

audio = {"data": Path("ami-es2004b-carphone-warehouse.wav").read_bytes(), "format": "wav"}
result = client.extract(
    "nvidia/parakeet-tdt-0.6b-v3",
    {"audio": audio},
)
print(result["data"])
```

## Next to the other speech models

| Model | LibriSpeech, 256 clean clips, word error rate | Time per request | Serverless price | Size |
|---|---|---|---|---|
| [Parakeet TDT 0.6B v3](/models/nvidia-parakeet-tdt-0-6b-v3.md) `nvidia/parakeet-tdt-0.6b-v3` | – | – | $0.00075 per audio minute | 627M |
| [whisper large v3 turbo](/models/openai-whisper-large-v3-turbo.md) `openai/whisper-large-v3-turbo` | 1.7% | – | $0.00276 per audio minute | 809M |

– = not measured / not timed / not listed.

[Both speech models](/models#speech)

## Facts

- Size: 627M parameters
- License: cc-by-4.0
- Reads: Audio
- Endpoint: extract

## Evaluation

- Accuracy: Run on SIE against the public dataset named in each figure
- Prices: Serverless list prices
- Sources: [Task page runs and their inputs](https://huggingface.co/datasets/superlinked/sie-task-evidence)